🚨 BREAKING: As the US Approves Nvidia H200 Sales, China Moves On Without Them
China’s Zhipu AI has just trained a serious, competitive multimodal AI model entirely on a domestic Chinese stack.
No Nvidia, no US chips and no TensorFlow or PyTorch.
GLM-Image was trained end-to-end on Huawei’s Ascend hardware, running on Huawei’s own MindSpore framework. Data prep, training, deployment, the lot.
That alone is HUGE. But the timing makes it even more perfect.
For years, the US strategy has been simple:
- choke off advanced chips
- slow China’s AI progress
- force long-term dependency
Yet here we are, proving yet again that China will not be held back.
Despite relentless sanctions, export controls, blacklists and political pressure, China has just proven, publicly, that high-quality, open-source multimodal AI does not require US silicon anymore. Bye bye!
This isn’t just some toy model wither, lets take a look:
Zhipu says GLM-Image delivers industry-leading performance among open-source models for text rendering and it’s especially strong with Chinese character generation, something Western models still struggle with. That matters when you’re building AI for Chinese-language reality, not English-first benchmarks.
Yes, it still trails ByteDance’s proprietary Seedream 4.5. But that’s almost beside the point. The significance isn’t “who’s number one”; it’s who controls the stack.
On the same day this news breaks, Washington approves exports of Nvidia’s H200 to China, after months of lobbying. The message from the US is basically:
“We’ll allow some access, under limits, to protect market share.”
And Beijing’s response? No thanks! Which is hilarious…
Customs have reportedly been told to block imports. Chinese firms have been told approvals will be “exceptional”. In other words: we’re done building our future around your permissions.
This isn’t China being “anti-Nvidia”, it’s China being post-Nvidia and self-sufficient.
For years, Nvidia dominated because there was no alternative and sanctions were supposed to freeze China in place. Instead, they forced a full-stack rebuild, chips, servers, frameworks, tooling, talent. That’s the China difference!
Once domestic stacks reach “good enough”, the dependency breaks permanently and once that happens, US leverage evaporates.
Ironically, approving the H200 now looks less like control and more like being late to a door that’s already closing or late to dinner, depending on how you look at it.
The real question now isn’t whether China can do advanced AI without US chips, that’s been answered. It’s how quickly the rest of the ecosystem follows.
What do you think hits first, large-scale adoption of domestic stacks, or a quiet global realisation that the chip war has already backfired?